Move a handful of channel messages to notify-without-persist
authorMatt Corallo <git@bluematt.me>
Thu, 24 Aug 2023 19:36:58 +0000 (19:36 +0000)
committerMatt Corallo <git@bluematt.me>
Tue, 12 Sep 2023 19:06:34 +0000 (19:06 +0000)
Many channel related messages don't actually change the channel
state in a way that changes the persisted channel. For example,
an `update_add_htlc` or `update_fail_htlc` message simply adds the
change to a queue, changing the channel state when we receive a
`commitment_signed` message.

In these cases there's really no reason to wake the background
processor at all - there's no response message and there's no state
update. However, note that if we close the channel we should
persist the `ChannelManager`. If we send an error message without
closing the channel, we should wake the background processor
without persisting.

Here we move to the appropriate `NotifyOption` on some of the
simpler channel message handlers.

lightning/src/ln/channelmanager.rs

index e97d6a5ee4289b7301e7b41edcd933eaa19f560c..eeeab90a8a7334b6cefb7ef8df57a5f245615d4c 100644 (file)
@@ -494,6 +494,10 @@ impl MsgHandleErrInternal {
                        channel_capacity: None,
                }
        }
+
+       fn closes_channel(&self) -> bool {
+               self.chan_id.is_some()
+       }
 }
 
 /// We hold back HTLCs we intend to relay for a random interval greater than this (see
@@ -1238,6 +1242,12 @@ struct PersistenceNotifierGuard<'a, F: Fn() -> NotifyOption> {
 }
 
 impl<'a> PersistenceNotifierGuard<'a, fn() -> NotifyOption> { // We don't care what the concrete F is here, it's unused
+       /// Notifies any waiters and indicates that we need to persist, in addition to possibly having
+       /// events to handle.
+       ///
+       /// This must always be called if the changes included a `ChannelMonitorUpdate`, as well as in
+       /// other cases where losing the changes on restart may result in a force-close or otherwise
+       /// isn't ideal.
        fn notify_on_drop<C: AChannelManager>(cm: &'a C) -> PersistenceNotifierGuard<'a, impl Fn() -> NotifyOption> {
                Self::optionally_notify(cm, || -> NotifyOption { NotifyOption::DoPersist })
        }
@@ -2152,9 +2162,14 @@ macro_rules! process_events_body {
                                processed_all_events = false;
                        }
 
-                       if result == NotifyOption::DoPersist {
-                               $self.needs_persist_flag.store(true, Ordering::Release);
-                               $self.event_persist_notifier.notify();
+                       match result {
+                               NotifyOption::DoPersist => {
+                                       $self.needs_persist_flag.store(true, Ordering::Release);
+                                       $self.event_persist_notifier.notify();
+                               },
+                               NotifyOption::SkipPersistHandleEvents =>
+                                       $self.event_persist_notifier.notify(),
+                               NotifyOption::SkipPersistNoEvents => {},
                        }
                }
        }
@@ -5560,6 +5575,8 @@ where
        }
 
        fn internal_open_channel(&self, counterparty_node_id: &PublicKey, msg: &msgs::OpenChannel) -> Result<(), MsgHandleErrInternal> {
+               // Note that the ChannelManager is NOT re-persisted on disk after this, so any changes are
+               // likely to be lost on restart!
                if msg.chain_hash != self.genesis_hash {
                        return Err(MsgHandleErrInternal::send_err_msg_no_close("Unknown genesis block hash".to_owned(), msg.temporary_channel_id.clone()));
                }
@@ -5659,6 +5676,8 @@ where
        }
 
        fn internal_accept_channel(&self, counterparty_node_id: &PublicKey, msg: &msgs::AcceptChannel) -> Result<(), MsgHandleErrInternal> {
+               // Note that the ChannelManager is NOT re-persisted on disk after this, so any changes are
+               // likely to be lost on restart!
                let (value, output_script, user_id) = {
                        let per_peer_state = self.per_peer_state.read().unwrap();
                        let peer_state_mutex = per_peer_state.get(counterparty_node_id)
@@ -5819,6 +5838,8 @@ where
        }
 
        fn internal_channel_ready(&self, counterparty_node_id: &PublicKey, msg: &msgs::ChannelReady) -> Result<(), MsgHandleErrInternal> {
+               // Note that the ChannelManager is NOT re-persisted on disk after this (unless we error
+               // closing a channel), so any changes are likely to be lost on restart!
                let per_peer_state = self.per_peer_state.read().unwrap();
                let peer_state_mutex = per_peer_state.get(counterparty_node_id)
                        .ok_or_else(|| {
@@ -5997,6 +6018,9 @@ where
                //encrypted with the same key. It's not immediately obvious how to usefully exploit that,
                //but we should prevent it anyway.
 
+               // Note that the ChannelManager is NOT re-persisted on disk after this (unless we error
+               // closing a channel), so any changes are likely to be lost on restart!
+
                let decoded_hop_res = self.decode_update_add_htlc_onion(msg);
                let per_peer_state = self.per_peer_state.read().unwrap();
                let peer_state_mutex = per_peer_state.get(counterparty_node_id)
@@ -6078,6 +6102,8 @@ where
        }
 
        fn internal_update_fail_htlc(&self, counterparty_node_id: &PublicKey, msg: &msgs::UpdateFailHTLC) -> Result<(), MsgHandleErrInternal> {
+               // Note that the ChannelManager is NOT re-persisted on disk after this (unless we error
+               // closing a channel), so any changes are likely to be lost on restart!
                let per_peer_state = self.per_peer_state.read().unwrap();
                let peer_state_mutex = per_peer_state.get(counterparty_node_id)
                        .ok_or_else(|| {
@@ -6101,6 +6127,8 @@ where
        }
 
        fn internal_update_fail_malformed_htlc(&self, counterparty_node_id: &PublicKey, msg: &msgs::UpdateFailMalformedHTLC) -> Result<(), MsgHandleErrInternal> {
+               // Note that the ChannelManager is NOT re-persisted on disk after this (unless we error
+               // closing a channel), so any changes are likely to be lost on restart!
                let per_peer_state = self.per_peer_state.read().unwrap();
                let peer_state_mutex = per_peer_state.get(counterparty_node_id)
                        .ok_or_else(|| {
@@ -7476,8 +7504,21 @@ where
        L::Target: Logger,
 {
        fn handle_open_channel(&self, counterparty_node_id: &PublicKey, msg: &msgs::OpenChannel) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-               let _ = handle_error!(self, self.internal_open_channel(counterparty_node_id, msg), *counterparty_node_id);
+               // Note that we never need to persist the updated ChannelManager for an inbound
+               // open_channel message - pre-funded channels are never written so there should be no
+               // change to the contents.
+               let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || {
+                       let res = self.internal_open_channel(counterparty_node_id, msg);
+                       let persist = match &res {
+                               Err(e) if e.closes_channel() => {
+                                       debug_assert!(false, "We shouldn't close a new channel");
+                                       NotifyOption::DoPersist
+                               },
+                               _ => NotifyOption::SkipPersistHandleEvents,
+                       };
+                       let _ = handle_error!(self, res, *counterparty_node_id);
+                       persist
+               });
        }
 
        fn handle_open_channel_v2(&self, counterparty_node_id: &PublicKey, msg: &msgs::OpenChannelV2) {
@@ -7487,8 +7528,13 @@ where
        }
 
        fn handle_accept_channel(&self, counterparty_node_id: &PublicKey, msg: &msgs::AcceptChannel) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-               let _ = handle_error!(self, self.internal_accept_channel(counterparty_node_id, msg), *counterparty_node_id);
+               // Note that we never need to persist the updated ChannelManager for an inbound
+               // accept_channel message - pre-funded channels are never written so there should be no
+               // change to the contents.
+               let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || {
+                       let _ = handle_error!(self, self.internal_accept_channel(counterparty_node_id, msg), *counterparty_node_id);
+                       NotifyOption::SkipPersistHandleEvents
+               });
        }
 
        fn handle_accept_channel_v2(&self, counterparty_node_id: &PublicKey, msg: &msgs::AcceptChannelV2) {
@@ -7508,8 +7554,19 @@ where
        }
 
        fn handle_channel_ready(&self, counterparty_node_id: &PublicKey, msg: &msgs::ChannelReady) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-               let _ = handle_error!(self, self.internal_channel_ready(counterparty_node_id, msg), *counterparty_node_id);
+               // Note that we never need to persist the updated ChannelManager for an inbound
+               // channel_ready message - while the channel's state will change, any channel_ready message
+               // will ultimately be re-sent on startup and the `ChannelMonitor` won't be updated so we
+               // will not force-close the channel on startup.
+               let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || {
+                       let res = self.internal_channel_ready(counterparty_node_id, msg);
+                       let persist = match &res {
+                               Err(e) if e.closes_channel() => NotifyOption::DoPersist,
+                               _ => NotifyOption::SkipPersistHandleEvents,
+                       };
+                       let _ = handle_error!(self, res, *counterparty_node_id);
+                       persist
+               });
        }
 
        fn handle_shutdown(&self, counterparty_node_id: &PublicKey, msg: &msgs::Shutdown) {
@@ -7523,8 +7580,19 @@ where
        }
 
        fn handle_update_add_htlc(&self, counterparty_node_id: &PublicKey, msg: &msgs::UpdateAddHTLC) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-               let _ = handle_error!(self, self.internal_update_add_htlc(counterparty_node_id, msg), *counterparty_node_id);
+               // Note that we never need to persist the updated ChannelManager for an inbound
+               // update_add_htlc message - the message itself doesn't change our channel state only the
+               // `commitment_signed` message afterwards will.
+               let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || {
+                       let res = self.internal_update_add_htlc(counterparty_node_id, msg);
+                       let persist = match &res {
+                               Err(e) if e.closes_channel() => NotifyOption::DoPersist,
+                               Err(_) => NotifyOption::SkipPersistHandleEvents,
+                               Ok(()) => NotifyOption::SkipPersistNoEvents,
+                       };
+                       let _ = handle_error!(self, res, *counterparty_node_id);
+                       persist
+               });
        }
 
        fn handle_update_fulfill_htlc(&self, counterparty_node_id: &PublicKey, msg: &msgs::UpdateFulfillHTLC) {
@@ -7533,13 +7601,35 @@ where
        }
 
        fn handle_update_fail_htlc(&self, counterparty_node_id: &PublicKey, msg: &msgs::UpdateFailHTLC) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-               let _ = handle_error!(self, self.internal_update_fail_htlc(counterparty_node_id, msg), *counterparty_node_id);
+               // Note that we never need to persist the updated ChannelManager for an inbound
+               // update_fail_htlc message - the message itself doesn't change our channel state only the
+               // `commitment_signed` message afterwards will.
+               let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || {
+                       let res = self.internal_update_fail_htlc(counterparty_node_id, msg);
+                       let persist = match &res {
+                               Err(e) if e.closes_channel() => NotifyOption::DoPersist,
+                               Err(_) => NotifyOption::SkipPersistHandleEvents,
+                               Ok(()) => NotifyOption::SkipPersistNoEvents,
+                       };
+                       let _ = handle_error!(self, res, *counterparty_node_id);
+                       persist
+               });
        }
 
        fn handle_update_fail_malformed_htlc(&self, counterparty_node_id: &PublicKey, msg: &msgs::UpdateFailMalformedHTLC) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-               let _ = handle_error!(self, self.internal_update_fail_malformed_htlc(counterparty_node_id, msg), *counterparty_node_id);
+               // Note that we never need to persist the updated ChannelManager for an inbound
+               // update_fail_malformed_htlc message - the message itself doesn't change our channel state
+               // only the `commitment_signed` message afterwards will.
+               let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || {
+                       let res = self.internal_update_fail_malformed_htlc(counterparty_node_id, msg);
+                       let persist = match &res {
+                               Err(e) if e.closes_channel() => NotifyOption::DoPersist,
+                               Err(_) => NotifyOption::SkipPersistHandleEvents,
+                               Ok(()) => NotifyOption::SkipPersistNoEvents,
+                       };
+                       let _ = handle_error!(self, res, *counterparty_node_id);
+                       persist
+               });
        }
 
        fn handle_commitment_signed(&self, counterparty_node_id: &PublicKey, msg: &msgs::CommitmentSigned) {
@@ -7553,8 +7643,19 @@ where
        }
 
        fn handle_update_fee(&self, counterparty_node_id: &PublicKey, msg: &msgs::UpdateFee) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-               let _ = handle_error!(self, self.internal_update_fee(counterparty_node_id, msg), *counterparty_node_id);
+               // Note that we never need to persist the updated ChannelManager for an inbound
+               // update_fee message - the message itself doesn't change our channel state only the
+               // `commitment_signed` message afterwards will.
+               let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || {
+                       let res = self.internal_update_fee(counterparty_node_id, msg);
+                       let persist = match &res {
+                               Err(e) if e.closes_channel() => NotifyOption::DoPersist,
+                               Err(_) => NotifyOption::SkipPersistHandleEvents,
+                               Ok(()) => NotifyOption::SkipPersistNoEvents,
+                       };
+                       let _ = handle_error!(self, res, *counterparty_node_id);
+                       persist
+               });
        }
 
        fn handle_announcement_signatures(&self, counterparty_node_id: &PublicKey, msg: &msgs::AnnouncementSignatures) {